The effect of personalized orthopedic insoles on plantar pressure during running in subtle cavus foot.

Objective: This study aims to investigate the patterns of plantar pressure distribution during running for patients with subtle cavus foot (SCF) and determine the impact of personalized orthopedic insoles with forefoot wedge on plantar pressure distribution in patients with SCF. Methods: Sixteen undergraduate participants (8 with SCF and 8 with normal arches) were recruited based on arch height index measurements. Two full-length insoles were personalized for each SCF based on plantar pressures during running, an arch support insole (ASI) and an arch support with forefoot wedge insole (AFI). Foot pressure data collected during different insole conditions in running, analyzing ten regions of footprints for peak pressure and pressure-time integral. Results: Higher peak pressures were observed in patients with SCF at the medial forefoot (p = 0.021), medial heel (p = 0.013), and lateral heel (p = 0.025), with a higher pressure-time integral also noted at the medial forefoot (p = 0.025), medial heel (p = 0.015), and lateral heel (p = 0.047) when compared to normal arches. Compared with without-insole, both the AFI and the ASI reduced peak pressure at the medial (AFI p = 0.011; ASI p = 0.024) and lateral heel (AFI p = 0.028; ASI p = 0.032). The AFI reduced peak pressure at the medial heel (p = 0.013) compared with the ASI. Both the AFI and the ASI reduced pressure-time integral at the medial forefoot (AFI p = 0.003; ASI p = 0.026), central forefoot (AFI p = 0.005; ASI p = 0.011), medial heel (AFI p = 0.017; ASI p = 0.005), and lateral heel (AFI p = 0.017; ASI p = 0.019). Additionally, the ASI reduced pressure-time integral at the big toe (p = 0.015) compared with the without-insole. Conclusion: These findings demonstrate that during running in patients with SCF, plantar pressures are concentrated in the forefoot and heel compared to the normal arch. The personalized orthotic insoles can be used to effectively redistribute plantar pressure in patients with SCF running. Incorporating a forefoot wedge to specifically address the biomechanical abnormalities associated with SCF may enhance the effectiveness of orthopedic insoles.

Comparative Analysis of the Diagonal Stride Technique during Roller Skiing and On-Snow Skiing in Youth Cross-Country Skiers.

Roller skiing is one primary form of training method as it is an off-snow equivalent to cross-country (XC) skiing during the dry land preseason training, but the results could only be applied to on-snow skiing with appropriate caution. The aim of this present study was to investigate the similarities and differences in roller skiing and on-snow skiing with the diagonal stride (DS) technique. Six youth (age: 14.3 ± 2.9 years) skiers participated in this study. Two high-definition video camcorders and FastMove 3D Motion 2.23.3.3101 were used to obtain the three-dimensional kinematic data. The cycle characteristics and joint angle ROM of the DS technique while skiing on different surfaces were similar. Almost all joint angle-time curves that were obtained from roller skiing showed a moderate-to-high degree of similarity to the angle-time curves obtained from on-snow skiing, except the hip adduction-abduction angle. The differences between roller skiing and on-snow skiing were mainly found in the body and calf anteversion angles, and the joint angles at critical instants. DS roller skiing can simulate DS on-snow skiing to a large extent in youth athletes. The hip movement, knee flexion, and calf anteversion at ski/roller ski touchdown and take-off, pole inclination at pole touchdown, body anteversion angle, and trunk anteversion angle at pole touchdown were the points that required caution when transferring preseason practice roller skiing to on-snow skiing.

Effects of step lengths on biomechanical characteristics of lower extremity during split squat movement.

Objective: To quantify the effects of increasing the step length of the split squat on changes in kinematics, kinetics, and muscle activation of the lower extremity. Methods: Twenty male college students participated in the test (age: 23.9 ± 3.7, height: 175.1 ± 4.9). Data on kinematics, kinetics, and EMG were collected during split squat exercise at four different step lengths in a non-systematic manner. One-way repeated measurements ANOVA were used to compare characteristic variables of peak angle, moment, and RMS among the four step length conditions. Results: The step length significantly changes the peak angles of the hip (p = 0.011), knee (p = 0.001), ankle (p < 0.001) joint, and the peak extension moment of the hip (p < 0.001), knee (p = 0.002) joint, but does not affect the ankle peak extension moment (p = 0.357) during a split squat. Moreover, a significant difference was observed in the EMG of gluteus maximus (p < 0.001), vastus medialis (p = 0.013), vastus lateralis (p = 0.020), biceps femoris (p = 0.003), Semitendinosus (p < 0.001), medialis gastrocnemius (p = 0.035) and lateralis gastrocnemius (p = 0.005) during four step lengths, but no difference in rectus femoris (p = 0.16). Conclusion: Increases in step length of split squat had a greater activation on the hip extensor muscles while having a limited impact on the knee extensor muscles. The ROM, joint moment, and muscle activation of the lead limb in the split squat all should be considered in cases of individual preventative or rehabilitative prescription of the exercise. Moreover, the optimal step length for strength training in healthy adults appears to be more suitable when it is equal to the length of the individual lower extremity.

PME: pruning-based multi-size embedding for recommender systems.

Embedding is widely used in recommendation models to learn feature representations. However, the traditional embedding technique that assigns a fixed size to all categorical features may be suboptimal due to the following reasons. In recommendation domain, the majority of categorical features' embeddings can be trained with less capacity without impacting model performance, thereby storing embeddings with equal length may incur unnecessary memory usage. Existing work that tries to allocate customized sizes for each feature usually either simply scales the embedding size with feature's popularity or formulates this size allocation problem as an architecture selection problem. Unfortunately, most of these methods either have large performance drop or incur significant extra time cost for searching proper embedding sizes. In this article, instead of formulating the size allocation problem as an architecture selection problem, we approach the problem from a pruning perspective and propose Pruning-based Multi-size Embedding (PME) framework. During the search phase, we prune the dimensions that have the least impact on model performance in the embedding to reduce its capacity. Then, we show that the customized size of each token can be obtained by transferring the capacity of its pruned embedding with significant less search cost. Experimental results validate that PME can efficiently find proper sizes and hence achieve strong performance while significantly reducing the number of parameters in the embedding layer.

Auto-GNN: Neural architecture search of graph neural networks.

Graph neural networks (GNNs) have been widely used in various graph analysis tasks. As the graph characteristics vary significantly in real-world systems, given a specific scenario, the architecture parameters need to be tuned carefully to identify a suitable GNN. Neural architecture search (NAS) has shown its potential in discovering the effective architectures for the learning tasks in image and language modeling. However, the existing NAS algorithms cannot be applied efficiently to GNN search problem because of two facts. First, the large-step exploration in the traditional controller fails to learn the sensitive performance variations with slight architecture modifications in GNNs. Second, the search space is composed of heterogeneous GNNs, which prevents the direct adoption of parameter sharing among them to accelerate the search progress. To tackle the challenges, we propose an automated graph neural networks (AGNN) framework, which aims to find the optimal GNN architecture efficiently. Specifically, a reinforced conservative controller is designed to explore the architecture space with small steps. To accelerate the validation, a novel constrained parameter sharing strategy is presented to regularize the weight transferring among GNNs. It avoids training from scratch and saves the computation time. Experimental results on the benchmark datasets demonstrate that the architecture identified by AGNN achieves the best performance and search efficiency, comparing with existing human-invented models and the traditional search methods.

Subarchitecture Ensemble Pruning in Neural Architecture Search.

Neural architecture search (NAS) is gaining more and more attention in recent years because of its flexibility and remarkable capability to reduce the burden of neural network design. To achieve better performance, however, the searching process usually costs massive computations that might not be affordable for researchers and practitioners. Although recent attempts have employed ensemble learning methods to mitigate the enormous computational cost, however, they neglect a key property of ensemble methods, namely diversity, which leads to collecting more similar subarchitectures with potential redundancy in the final design. To tackle this problem, we propose a pruning method for NAS ensembles called " subarchitecture ensemble pruning in neural architecture search (SAEP)." It targets to leverage diversity and to achieve subensemble architectures at a smaller size with comparable performance to ensemble architectures that are not pruned. Three possible solutions are proposed to decide which subarchitectures to prune during the searching process. Experimental results exhibit the effectiveness of the proposed method by largely reducing the number of subarchitectures without degrading the performance.

On Robustness of Neural Architecture Search Under Label Noise.

Neural architecture search (NAS), which aims at automatically seeking proper neural architectures given a specific task, has attracted extensive attention recently in supervised learning applications. In most real-world situations, the class labels provided in the training data would be noisy due to many reasons, such as subjective judgments, inadequate information, and random human errors. Existing work has demonstrated the adverse effects of label noise on the learning of weights of neural networks. These effects could become more critical in NAS since the architectures are not only trained with noisy labels but are also compared based on their performances on noisy validation sets. In this paper, we systematically explore the robustness of NAS under label noise. We show that label noise in the training and/or validation data can lead to various degrees of performance variations. Through empirical experiments, using robust loss functions can mitigate the performance degradation under symmetric label noise as well as under a simple model of class conditional label noise. We also provide a theoretical justification for this. Both empirical and theoretical results provide a strong argument in favor of employing the robust loss function in NAS under high-level noise.